Streaming Data Integration: Challenges and Opportunities. Nesime Tatbul

Transcription

1 Streaming Data Integration: Challenges and Opportunities Nesime Tatbul

2 Talk Outline Integrated data stream processing An example project: MaxStream Architecture Query model Conclusions ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 2

3 Data Stream Processing Monitoring applications require collecting, processing, disseminating, and reacting to real-time events from push-based data sources. Store and Pull model of traditional databases does not work well. Query DBMS Answer Data SPE Answer Data Base Query Base Traditional Database Systems Stream Processing Engines ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 3

4 Integrated Data Stream Processing Today, integration support for SPEs is needed in three main forms: 1. across multiple streaming data sources example: news feeds, weather sensors, traffic cameras 2. over multiple SPEs example: supply-chain management 3. between SPEs and traditional DBMSs example: operational business intelligence ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 4

5 #1: Streaming Data Source Integration Goal: Integrated querying over multiple, potentially heterogeneous streaming data sources Challenges: Schemas of different sources can differ from one another and from the input schemas of the already running CQs. Input sources or the network can introduce imperfections into the stream. Adapters may become a bottleneck. Current state of the art: Commercial SPEs offer a collection of common adapters and SDKs for developing custom ones. Mapping Data to Queries [Hentschel et al] ASPEN project [Ives et al] ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 5

6 #2: SPE-SPE Integration Goal: Integrated querying over multiple, potentially heterogeneous SPEs to exploit the advantages of distributed operation to exploit specialized capabilities and strengths of SPEs to provide higher-level monitoring over large-scale enterprises with loosely-coupled operational units Challenges: The need for functional integration The need to deal with heterogeneity at different levels (e.g., query models, capabilities, performance, interfaces) Current state of the art: MaxStream project [Tatbul et al] ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 6

7 #3: SPE-DBMS Integration Goal: Integrated querying over SPEs and traditional database systems Challenges: Bridge the data vs. operation gap between the two worlds. Find the right language and architecture primitives for the required level of querying, persistence, and performance. Current state of the art: Languages [STREAM CQL, StreamSQL] Architectures SPE-based [typical SPEs such as Coral8, StreamBase] DBMS-based * stream-relational systems such as TelegraphCQ/Truviso, DataCell, DejaVu, MaxStream] ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 7

8 MaxStream: A Platform for SPE-SPE and SPE-DBMS Integration Client Application Federation Layer Data Wrapper Agent Wrapper Wrapper Key design ideas: Uniform query language and API Relational database infrastructure as the basis for the federation layer (in our case: SAP MaxDB and SAP MaxDB Federator) DB DB SPE SPE Just enough streaming capability inside the federation layer ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 8

9 MaxStream vs. Traditional Virtual Integration Traditional Virtual Integration Goal: to query across multiple autonomous & heterogeneous data sources Source wrappers send queries and receive answers Sources host the data Mapping between source schemas and a global schema Queries are posed against the global schema More focus on data locality MaxStream Goal: to query across multiple autonomous & heterogeneous SPEs (and DBMSs) SPE wrappers send queries and data, and receive answers SPEs host the CQs Mapping between SPE CQ models and a global CQ model CQs are posed and data is fed against the global CQ model More focus on functional heterogeneity ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 9

10 Output Events MaxStream Architecture Client Application Output Events Input Events DDL/DML statements in MaxStream s SQL Dialect Metadata Input Event Tables Output Event Tables SQL Parser Query Rewriter Query Optimizer Query Executer MaxStream Federator SQL Dialect Translator Input Events Data Agent for SPE DDL/DML statements in SPE s SQL Dialect Data Data Agent Agent Data Agent for SPE SPE s SDK SPE s SDK MaxDB ODBC DB DB MaxDB ODBC SPE ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich SPE 10

11 MaxStream Architecture Two Key Building Blocks Streaming inputs through MaxStream ISTREAM operator for persistent input events Tuple queues for transient input events Streaming outputs through MaxStream Monitoring Select operator over event tables Persistent event tables for persistent output events In-memory event tables for transient output events ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 11

12 Streaming Input Events The ISTREAM ( Insert STREAM ) Operator Inspired by the relation-to-stream operator of the same name in the STREAM Project, that streams new tuples being inserted into a given relation. Example: OrdersTable(ClientId, OrderId, ProductId, Quantity) INSERT INTO STREAM OrdersStream SELECT ClientId, OrderId, ProductId, Quantity FROM ISTREAM(OrdersTable); T r1 r2 r3 ICDE NTII Workshop, 2010 T+1 r1 r2 r3 r4 r5 ISTREAM(OrdersTable) at T+1 returns: <r4, T+1>, <r5, T+1> Nesime Tatbul, ETH Zurich 12

13 Streaming Output Events Opposite of streaming input events, but Unlike the SPE interface, the client application interface is not push-based. The Monitoring Select Operator Select operation blocks until there is at least one row to return. For continuous monitoring, the client program re-issues Monitoring Select in a loop. Monitoring Select operates on event tables. Example: Detect unusually large order volumes. SELECT * FROM /*+ EVENT */ TotalSalesTable WHERE TotalSales > ; ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 13

14 ISTREAM and Monitoring Select in Action Data Feeder Client // Continuous Insert into OrdersTable kept in MaxStream CREATE TABLE OrdersTable; WHILE (true) { INSERT INTO OrdersTable VALUES ( ); sleep(period); }; // Continuous Insert into OrdersStream kept in the SPE CREATE STREAM OrdersStream; INSERT INTO STREAM OrdersStream SELECT FROM ISTREAM(OrdersTable); Monitoring Clients // Setting up the Continuous Query to push down to the SPE INSERT INTO STREAM TotalSalesStream SELECT SUM( ) FROM OrdersStream KEEP 1 HOUR; // Streaming Output Events inserted by the SPE CREATE TABLE TotalSalesTable; // Sales spikes Query to run in MaxStream WHILE (true) { SELECT * FROM /*+EVENT*/ TotalSalesTable WHERE TotalSales > ; }; MaxStream Monitoring Select SPE TotalSalesTable TotalSalesStream OrdersTable ISTREAM OrdersStream INSERT INTO STREAM TotalSalesStream SELECT SUM( ) FROM OrdersStream KEEP 1 HOUR; ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 14

15 Hybrid Queries in MaxStream Hybrid queries are continuous queries that join Streams with Tables. Two important factors that affect efficiency: The streaming data source must be first in the join ordering. Hybrid queries can be rewritten to perform the join within the MaxStream Federator, removing the need for the SPE to establish connections to external databases. One can conveniently use hybrid queries in MaxStream in two ways: To enrich the input stream before it is passed to the SPE To enrich the output stream after it is received from the SPE ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 15

16 MaxStream Query Model Problem: Heterogeneity of SPE query models Syntax heterogeneity Language clauses/keywords for common constructs syntactically differ. Capability heterogeneity Support for certain query types differs. Execution model heterogeneity Underlying query execution models differ. Not exposed to the application developer at the language syntax level. First step towards a solution: Create a model to analyze and predict the query execution semantics of SPEs. ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 16

17 The SECRET Model What affects query results produced by an SPE? ScopE: Given a query with certain window properties, what are the potential window intervals? Content: Given an input stream, what are the actual contents for those window intervals? REport: Under what conditions, do those window contents become visible to the query processor for evaluation? Tick: What drives an SPE to take action on a given input stream? Query Result = F(system, query, input) Tick REport ScopE Content ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 17

18 The SECRET of a Query Plan SECRET SECRET ScopE window Tick REport Content query operator ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 18

19 The SECRET of an SPE Tick: tuple-driven (e.g., Aurora, Borealis, StreamBase, TelegraphCQ, Truviso) time-driven (e.g., STREAM, Oracle CEP) batch-driven (e.g., Coral8, *Jain et al, VLDB 08+) REport: window close & non-empty (e.g., StreamBase) content change & non-empty (e.g., Coral8) window close & content change & non-empty (e.g., STREAM) ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 19

20 MaxStream: Future Outlook Query model how to extend SECRET (other query types, analysis of other SPEs, input imperfections) how to use SECRET in MaxStream ( SECRET-based query and SPE capability analyzer) Capability- and Cost-based query optimization Transactional stream processing Distributed operation ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 20

21 Conclusions Today, integration support for SPEs is needed in three main forms: across sources, SPE-SPE, SPE-DBMS. There are many open research challenges. MaxStream takes up some of these challenges for SPE-SPE and SPE-DBMS integration. More information about MaxStream: ICDE NTII Workshop, 2010 Nesime Tatbul, ETH Zurich 21